DECOMPOSITION OF TENSOR PRODUCTS OF MODULAR IRREDUCIBLE REPRESENTATIONS FOR SL 3 (WITH AN APPENDIX BY C.M. RINGEL)

Transcription

1 DECOMPOSITION OF TENSOR PRODUCTS OF MODULAR IRREDUCIBLE REPRESENTATIONS FOR SL 3 (WITH AN APPENDIX BY CM RINGEL) C BOWMAN, SR DOTY, AND S MARTIN Abstract We give an algorithm for working out the indecomposable direct summands in a Krull Schmidt decomposition of a tensor product of two simple modules for G = SL 3 in characteristics 2 and 3 It is shown that there is a finite family of modules such that every such indecomposable summand is expressible as a twisted tensor product of members of that family Along the way we obtain the submodule structure of various Weyl and tilting modules Some of the tilting modules that turn up in characteristic 3 are not rigid; these seem to provide the first example of non-rigid tilting modules These non-rigid tilting modules lead to examples of non-rigid projective indecomposable modules for Schur algebras, as shown in the Appendix Higher characteristics (for SL 3) will be considered in a later paper 1 Introduction We begin by explaining our motivation, which may be formulated for an arbitrary semisimple algebraic group in positive characteristic 11 Let G be a semisimple, simply connected linear algebraic group over an algebraically closed field K of positive characteristic p We fix a Borel subgroup B and a maximal torus T with T B G and we let B determine the negative roots We write X = X(T ) for the character group of T and let X + denote the set of dominant weights By G-module we always mean a rational G-module, ie a K[G]-comodule, where K[G] is the coordinate algebra of G For each λ X + we have the following (see [14]) finite dimensional G- modules: L(λ) simple module of highest weight λ; (λ) Weyl module of highest weight λ; (λ) = ind G B K λ; dual Weyl module of highest weight λ; T(λ) indecomposable tilting module of highest weight λ where K λ is the 1-dimensional B-module upon which T acts by the character λ with the unipotent radical of B acting trivially The simple modules L(λ) are contravariantly selfdual The module (λ) has simple socle isomorphic to L(λ); the module (λ) is isomorphic to τ (λ), the contravariant dual of (λ), hence has simple head isomorphic to L(λ) The central problem which interests us is as follows Problem 1 Describe the indecomposable direct summands of an arbitrary tensor product of the form L(λ) L(µ), for λ, µ X + Date: 12 July Mathematics Subject Classification 20C30 1

2 2 C BOWMAN, SR DOTY, AND S MARTIN As usual, a superscript M [j] on a G-module M indicates that the structure has been twisted by the jth power of the Frobenius endomorphism on G By Steinberg s tensor product theorem, there are twisted tensor product factorizations L(λ) L(λ 0 ) L(λ 1 ) [1] L(λ 2 ) [2] ; L(µ) L(µ 0 ) L(µ 1 ) [1] L(µ 2 ) [2] where λ = λ j p j, µ = µ j p j are the p-adic expansions (unique) such that each λ j, µ j belongs to the restricted region X 1 = {ν X + α, ν p 1 for all simple roots α} Putting these factorizations into the original tensor product we obtain (111) L(λ) L(µ) j 0 ( L(λ 1 ) L(µ 1 ) ) [j] and thus we see that in Problem 1 one should first study the case where both highest weights in question are restricted Assume that Problem 1 has been solved for all pairs of restricted weights (note that this is a finite problem for any given G) Let F = F(G) be the set of isomorphism classes of indecomposable direct summands appearing in some L L, for a pair L, L of restricted simple G-modules Let [L L : I] be the multiplicity of I F as a direct summand of L L Then one can express each tensor product L(λ j ) L(µ j ) as a finite direct sum of indecomposable modules (112) L(λ j ) L(µ j ) I F [L(λ j ) L(µ j ) : I] I Thus, the original tensor product L(λ) L(µ) has a decomposition of the form L(λ) L(µ) j 0 I F [L(λj ) L(µ j ) : I] I [j] and by interchanging the order of the product and sum we obtain the decomposition (113) L(λ) L(µ) ( j 0 [L(λj ) L(µ j ) : I j ] ) j 0 I[j] j where the direct sum is taken over the set of all finite sequences (I 0, I 1, I 2, ) of members of F This gives a direct sum decomposition of L(λ) L(µ) in terms of twisted tensor products of modules in F If all such twisted tensor products are themselves indecomposable as G-modules, then we have in some sense solved Problem 1 for general λ, µ Even when this isn t true we have still obtained a first approximation towards a solution to Problem 1 This leads us to the following secondary set of problems: Problem 2 Given G, (a) classify the members of the family F = F(G) and compute the multiplicities [L L : I] for I F, L, L restricted; (b) determine conditions under which a twisted tensor product of members from F remains indecomposable; (c) determine the module structure of the members of F

3 DECOMPOSITION OF TENSOR PRODUCTS 3 Let G r denote the kernel of the rth iterate of the Frobenius, let G r T denote the inverse image of T under the same map, and let Q r (λ) denote the G r T -injective hull of L(λ) for any λ X r, where X r := {ν X + α, ν p r 1 for all simple roots α} Let h denote the Coxeter number of G If p 2h 2 then Q r (µ) has (for any µ X r ) a G-module structure; this structure is unique in the sense that any two such G-module structures are equivalent (These statements are expected to hold for all p) Concerning Problem 2(b) we observe the following Lemma Assume that p 2h 2 or that if p < 2h 2 then Q 1 (µ) has a unique G- module structure for all µ X 1 If each member of the sequence (I j ) j 0 (I j F) has simple G 1 T -socle with restricted highest weight then the twisted tensor product j 0 is indecomposable as a G-module Hence P is indecomposable I[j] j Proof By assumption the socle of I j is simple, as a G 1 T -module, hence has the form L(µ(j)) for some µ(j) X 1 Hence the module I j embeds in the G 1 T -injective hull Q 1 (µ(j)) of L(µ(j)), for each j, so P := I 0 I [1] 1 I m [m] embeds in Q := Q 1 (µ(0)) Q 1 (µ(1)) [1] Q 1 (µ(m)) [m] By [14, II1116 Remark 2] the module Q has a G-module structure and is isomorphic to Q r (µ), where µ = j µ(t j)p j Since Q r (µ) has simple G r T -socle L(µ) it follows that P also has simple G r T -socle L(µ), and thus has simple G-socle L(µ) We note that in Types A 1 and A 2 (G = SL 2, SL 3 ) it is known that Q 1 (µ) has a unique G-module structure for all µ X 1, for any p In the case G = SL 2 (studied in [8]) it turns out that for any p the members of F are always indecomposable tilting modules with simple G 1 T -socle of restricted highest weight, so the determination of the family F and the multiplicities [L L : I] leads in that case to a complete solution of Problem 1 for all pairs of dominant weights The purpose of this paper is to examine the next most complicated case, namely the case G = SL 3 In that case, we will see that all members of F have simple G 1 T -socle of restricted highest weight when p = 2, and this holds with only two exceptions when p = 3, so the decomposition (113) is decisive in characteristic 2 and provides a great deal of information in characteristic 3 Furthermore, although in characteristic 3 the summands in (113) are not always indecomposable, by analyzing the further splittings which arise, we show that there is a finite family F, closely related to F, such that every indecomposable direct summand of L(λ) L(µ) is isomorphic to a twisted tensor product of members of F Thus, we obtain a complete solution to Problem 1 in characteristics 2 and 3 12 The paper is organized as follows In Section 2 we recall known facts that we use Our main technique is to compute structure of certain Weyl modules (using a computer when necessary) and use that structure to deduce structural information on certain tilting modules The main results obtained by our computations are given in Sections 3 and 4 To be specific, the structure of the relevant Weyl modules is given in 31 and 41, while the main results on tensor products including description of the family F, multiplicities [L L : I] for restricted simples L, L and I F, and structure of members of F (in most cases) are summarized in 32 and 42 One will also find worked examples in those sections

4 4 C BOWMAN, SR DOTY, AND S MARTIN In characteristic 2 all members of F(SL 3 ) are tilting modules with simple G 1 T -socle of restricted highest weight, so the decomposition (113) gives a complete answer to Problem 1 for all pairs of dominant weights This is similar to what happens for G = SL 2 Moreover, each member of F(SL 3 ) in this case is rigid (a module is called rigid if its radical and socle filtrations coincide) and can be described by a strong diagram in the sense of [1] Recall that in [1] a module diagram is a directed graph depicting the radical series of the module, in such a way that vertices correspond to composition factors and edges to non-split extensions, and a strong diagram is one in which the diagram also determines the socle series (One should consult [1] for precise statements) Characteristic 3 is more complicated (As standard notation, we write (a, b) for a highest weight of the form aϖ 1 + bϖ 2 where ϖ 1, ϖ 2 are the usual fundamental weights) First, all but two of the members of F(SL 3 ) have simple G 1 T -socle of restricted highest weight The two exceptional cases are in fact simple modules of highest weights (5, 2) and (2, 5) that are not restricted, and so one is forced to consider possible further splitting of summands in (113), in cases where one or both of these modules appears in a twisted tensor product on the right hand side (This happens only if the tensor square of the Steinberg module occurs in some factor in the right hand side of (111)) In all cases those further splittings can be worked out; see Proposition 43 This leads to the finite family F discussed in the last paragraph of 11 Furthermore, in characteristic 3 it turns out that four members of F(SL 3 ) namely the tilting modules T(3, 3), T(4, 3), T(3, 4), and T(4, 4) are not rigid and do not have strong Alperin diagrams The structure of one of the simplest of these examples, T(4, 3), is analyzed in detail in the Appendix by CM Ringel, using different methods Although not itself projective, Ringel shows that T(4, 3) is a quotient of the corresponding projective indecomposable for an appropriate Schur algebra, and thus he produces an example of a non-rigid projective indecomposable module for that Schur algebra The other non-rigid modules are subject to a similar analysis Preliminary calculations indicate that members of F(SL 3 ) are again rigid in characteristics higher than 3 The observed anomalies in characteristic 3 are associated with the fact that some of the Weyl modules which turn up are too close to the upper wall of the lowest p 2 -alcove and thus have composition factors with multiplicity greater than 1 (Those multiplicities follow, eg from [7], from knowledge of composition factor multiplicities in baby Verma modules, which are well known in this case) The simple characters for SL 3 have been known since [12, 13] Our results overlap somewhat with [15], [16] although our methods are different and we push the calculations further Larger characteristics, for which some calculations become in a sense independent of p, will be treated in a future paper This paper has been circulating for some time in various forms, and since the first version was made available, the preprint [2] has appeared, in which further examples of non-rigid tilting modules are obtained 2 Preliminaries We recall some general facts that will be used in our calculations 21 Let us recall Pillen s Theorem [18, 2, Corollary A] (see also [3, (25) Theorem]) Write St r for the rth Steinberg module L((p r 1)ρ) = ((p r 1)ρ) = T((p r 1)ρ) Then

5 DECOMPOSITION OF TENSOR PRODUCTS 5 for λ X r the tilting module T(2(p r 1)ρ + w 0 λ) is isomorphic to the indecomposable G-component of St r L((p r 1)ρ + w 0 λ) containing the weight vectors of highest weight 2(p r 1)ρ + w 0 λ 22 In general the formal character of a tilting module is not known, although for SL 3 they can all be worked out The following general result of Donkin (see [5, Proposition 55]) computes the formal character of certain tilting modules Let λ, µ X + and assume that (λ, α0 ) p, where α 0 is the highest short root Then: (221) ch T((p 1)ρ + λ) = ch L((p 1)ρ) e(ν) and for any ν X +, (222) (T((p 1)ρ + λ + pµ) : (ν)) = where N(ν) = {ξ X + : ν + ρ p(ξ + ρ) W λ} ξ N(ν) ν W λ (T(µ) : (ξ)) 23 Another useful general fact (that will be used repeatedly) is the observation that tilting modules are contravariantly self-dual: (231) τ T(λ) T(λ) for all λ X + This is because (by [14, II213]) contravariant duality interchanges (µ) and (µ), so τ T(λ) is again indecomposable tilting, of the same highest weight 24 Finally, there is a twisted tensor product theorem for tilting modules, assuming that Donkin s conjecture [3, (22) Conjecture] is valid or that p 2h 2 (It is well known [14, II1116, Remark 2] that the conjecture is valid for all p in case G = SL 3 ) For our purposes, it is convenient to reformulate the tensor product theorem in the following form First we observe that, given λ X + satisfying the condition (241) λ, α p 1, for all simple roots α, there exist unique weights λ, µ such that (242) λ = λ + pµ, λ (p 1)ρ + X 1, µ X + This is easy to see: for each λ i in λ = λ i ϖ i where the ϖ i are the fundamental weights, express λ i (p 1) (uniquely) in the form λ i (p 1) = r i + ps i with 0 r i p 1 Then set λ = (p 1)ρ + r i ϖ i and µ = s i ϖ i Now by induction on m using (241) and (242) one shows that every λ X + has a unique expression in the form (243) λ = m j=0 a j(λ) p j with a 0 (λ),, a m 1 (λ) (p 1)ρ + X 1 and a m (λ), α < p 1 for at least one simple root α Given λ X +, express λ in the form (243) Assume Donkin s conjecture holds if p < 2h 2 Then there is an isomorphism of G-modules m (244) T(λ) T(a j (λ)) [j] j=0

6 6 C BOWMAN, SR DOTY, AND S MARTIN To prove this one uses induction and [14, Lemma IIE9] (which is a slight reformulation of [3, (21) Proposition]) 3 Results for p = 2 For the rest of the paper we take G = SL 3 Conventions: Dominant weights are written as ordered pairs (a, b) of non-negative integers; one should read (a, b) as an abbreviation for aϖ 1 + bϖ 2 where ϖ 1, ϖ 2 are the fundamental weights, defined by the condition ϖ i, α j = δ ij When describing module structure, we shall always identify a simple module L(λ) with its highest weight λ Whenever possible we will depict the structure by giving an Alperin diagram (see [1] for definitions) with edges directed downwards, except in the uniserial case, where we will write M = [L s, L s 1,, L 1 ] for a module M with unique composition series 0 = M 0 M 1 M s 1 M s = M such that L j M j /M j 1 is simple for each j 31 Structure of certain Weyl modules for p = 2 The results given below were computer generated, using GAP [9] code available on the second author s web page (Some cases are obtainable from [6]) The restricted region X 1 in this case consists of the weights of the form (a, b) with 0 a, b 1, and we have (0, 0) = L(0, 0), (1, 0) = L(1, 0), (0, 1) = L(0, 1), (1, 1) = L(1, 1) These are all tilting modules Thus it follows immediately that all the members of F are tilting The structure of the other Weyl modules we need is depicted below The uniserial modules have structure (2, 0) = [(2, 0), (0, 1)], (0, 2) = [(0, 2), (1, 0)] (3, 0) = [(3, 0), (0, 0)], (0, 3) = [(0, 3), (0, 0)] (2, 1) = [(2, 1), (0, 2), (1, 0)], (1, 2) = [(1, 2), (2, 0), (0, 1)] Finally, the structure of (2, 2) is given by the diagram (2, 2) = (2,2) (0,3) (3,0) (0,0) We worked these out using explicit calculations in the hyperalgebra, by methods similar to those of [11, 19] 32 Restricted tensor product decompositions for p = 2 The indecomposable decompositions of restricted tensor products for p = 2 is as follows (We omit any decomposition of the form L(λ) L(µ) where one of λ, µ is zero) There is an involution on G-modules which on weights is the map λ w 0 (λ), where w 0 is the longest element of the Weyl group (In Type A this comes from a graph automorphism of the Dynkin diagram) We refer to this involution as symmetry, and we will often omit calculations and results that can be obtained by symmetry from a calculation or result already given

7 Proposition Suppose p = 2 DECOMPOSITION OF TENSOR PRODUCTS 7 (a) The indecomposable direct summands of tensor products of non-trivial restricted simple SL 3 -modules are as follows: (1) L(1, 0) L(1, 0) T(2, 0); L(0, 1) L(0, 1) T(0, 2); (2) L(1, 0) L(0, 1) T(1, 1) T(0, 0); (3) L(1, 0) L(1, 1) T(2, 1); L(0, 1) L(1, 1) T(1, 2); (4) L(1, 1) L(1, 1) T(2, 2) 2T(1, 1) Thus the family F(SL 3 ) is in this case given by F = {T(a, b) : 0 a, b 2} (b) The structure of the uniserial members of F is given as follows: T(0, 0) = [(0, 0)]; T(1, 0) = [(1, 0)]; T(1, 1) = [(1, 1)]; T(2, 0) = [(0, 1), (2, 0), (0, 1)] The structure diagrams of T(2, 1), T(2, 2) are displayed below: (1,0) (0,2) (2,1) (1,0) (0,2) (1,0) (0,0) (0,3) (3,0) (0,0) (2,2) (0,0) (0,3) (3,0) (0,0) and the structure diagrams of T(0, 1), T(0, 2), and T(1, 2) are obtained by symmetry from cases already listed (c) Each member of F has simple G 1 T -socle (and head) with highest weight belonging to the restricted region X 1 The proof is given in 33 First we consider consequences and give some examples Recall that a dominant weight is called minuscule if the weights of the corresponding Weyl module form a single Weyl group orbit For G = SL 3 the minuscule weights are (0, 0), (1, 0), and (0, 1) Corollary Let p = 2 Given arbitrary dominant weights λ, µ write λ = λ j p j, µ = µ j p j with λ j, µ j X 1 for all j 0 (a) In the decomposition (113), every direct summand is indecomposable Hence the indecomposable direct summands of L(λ) L(µ) are expressible as a twisted tensor product of members of F Conversely, every twisted tensor product of members of F occurs in some L(λ) L(µ) (b) L(λ) L(µ) is indecomposable if and only if for each j 0 the unordered pair {λ j, µ j } is one of the cases {(1, 0), (1, 0)}, {(0, 1), (0, 1)}, {(1, 0), (1, 1)}, {(0, 1), (1, 1)} or one of λ j, µ j is the trivial weight (0, 0) (c) Let m be the maximum j such that at least one of λ j, µ j is non-zero Then L(λ) L(µ) is indecomposable tilting, isomorphic to T(λ + µ), if and only if: (i) for each 0 j m 1, one of λ j, µ j is minuscule and the other is the Steinberg weight (1, 1), and (ii) {λ m, µ m } is one of the cases listed in part (b) Proof Part (a) follows from (113) and Lemma 11 Part (b) follow from the proposition and the discussion preceding (113), which shows that each L(λ j ) L(µ j ) must be itself

8 8 C BOWMAN, SR DOTY, AND S MARTIN indecomposable in order for L(λ) L(µ) to be indecomposable Then we get part (c) from part (b) by applying Donkin s tensor product theorem (244) Examples (i) To illustrate the procedure in part (a) of the corollary, we work out a specific example: L(7, 2) L(6, 3) ( L(1, 0) L(0, 1) ) ( L(1, 1) L(1, 1) ) [1] ( L(1, 0) L(1, 0) ) [2] ( T(1, 1) T(0, 0) ) ( T(2, 2) 2T(1, 1) ) [1] T(2, 0) [2] T(13, 5) 2T(6, 2) [1] 2T(11, 3) 2T(5, 1) [1] In the calculation, the first line follows from Steinberg s tensor product theorem, the second is from the proposition, and to get the last line we applied Donkin s tensor product theorem (244), after interchanging the order of sums and products (ii) We have L(3, 0) L(3, 2) ( L(1, 0) L(1, 0) ) ( L(1, 0) L(1, 1) ) [1] T(2, 0) T(2, 1) [1], which is indecomposable but not tilting This illustrates the procedure in part (b) of the corollary (iii) We have L(3, 0) L(3, 1) ( L(1, 0) L(1, 1) ) ( L(1, 0) L(1, 0) ) [1] T(2, 1) T(2, 0) [1] T(6, 1), illustrating part (c) of the corollary (iv) It is not the case that every indecomposable tilting module occurs as a direct summand of some tensor product of two simple modules For instance, neither T(3, 0) nor T(0, 3) (both of which are uniserial of length 3) can appear as one of the indecomposable direct summands on the right hand side of (113) This follows from (244) More generally, this applies to any non-simple tilting module of the form T(a, b) with one of a, b equal to zero and the other greater than 2 33 We now consider the proof of Proposition 32 First we compute the composition factor multiplicities of the restricted tensor products Let χ p (λ) = L(λ) be the formal character of L(λ) Then: (1) χ p (1, 0) χ p (1, 0) = χ p (2, 0) + 2χ p (0, 1); (2) χ p (1, 0) χ p (0, 1) = χ p (1, 1) + χ p (0, 0); (3) χ p (1, 0) χ p (1, 1) = χ p (2, 1) + 2χ p (0, 2) + 3χ p (1, 0); (4) χ p (0, 1) χ p (0, 1) = χ p (0, 2) + 2χ p (1, 0); (5) χ p (0, 1) χ p (1, 1) = χ p (1, 2) + 2χ p (2, 0) + 3χ p (0, 1); (6) χ p (1, 1) χ p (1, 1) = χ p (2, 2) + 2χ p (0, 3) + 2χ p (3, 0) + 2χ p (1, 1) + 4χ p (0, 0) Since L(1, 0) = T(1, 0), it follows that L(1, 0) L(1, 0) is tilting It must have T(2, 0) as a direct summand by highest weight considerations But T(2, 0) is contravariantly self-dual with L(0, 1) in the socle, so it follows that L(0, 1) appears with multiplicity at least 2 as a composition factor of T(2, 0) Now character considerations force the structure to be given by L(1, 0) L(1, 0) T(2, 0) where T(2, 0) = [(0, 1), (2, 0), (0, 1)] By symmetry we also have where T(0, 2) = [(1, 0), (0, 2), (1, 0)] L(0, 1) L(0, 1) T(0, 2)

9 DECOMPOSITION OF TENSOR PRODUCTS 9 L(1, 0) L(0, 1) is tilting and has a direct summand isomorphic to T(1, 1) = L(1, 1) By character considerations it follows that there is one other indecomposable summand, namely T(0, 0) = L(0, 0) Hence L(1, 0) L(0, 1) T(0, 0) T(1, 1) L(1, 0) L(1, 1) is tilting and has a direct summand T(2, 1) Self-duality of T(2, 1) forces a copy of L(1, 0) at the top, extending L(0, 2) This, along with the structure of the Weyl modules and known Ext information forces the structure of T(2, 1) to be as given in the statement of Proposition 32(b), and also forces By symmetry we obtain also L(1, 0) L(1, 1) T(2, 1) L(0, 1) L(1, 1) T(1, 2) Finally, L(1, 1) L(1, 1) is tilting, with a direct summand isomorphic to T(2, 2) The highest weights of all simple composition factors of the tensor product are in the same linkage class, excepting (1, 1), which appears with multiplicity 2 So two copies of T(1, 1) split off Moreover, T(2, 2) has a submodule isomorphic to (2, 2), thus contains L(0, 0) in the socle This forces another copy of L(0, 0) at the top of T(2, 2), and this along with known Ext information and the structure of the Weyl modules forces the structure of T(2, 2) to be as given in Proposition 32(b), and also forces L(1, 1) L(1, 1) T(2, 2) 2T(1, 1) All the claims in Proposition 32(a), (b) are now clear It remains to verify the claim in (c) It is known that Donkin s conjecture holds for SL 3, so T((p 1)ρ+λ) is as a G 1 T -module isomorphic to Q 1 ((p 1)ρ + w 0 λ) for any λ X 1 Thus T(2, 1), T(1, 2), and T(2, 2) each has a simple G 1 T -socle of restricted highest weight For T(2, 0) and T(0, 2) one can argue by contradiction, using the fact [3, (15) Proposition] that truncation to an appropriate Levi subgroup L maps indecomposable tilting modules for G onto indecomposable tilting modules for L Thus T(2, 0) and T(0, 2) truncate to T(2) for L SL 2, which is known to have simple L 1 T -socle and length three If T(2, 0) or T(0, 2) did not have simple G 1 T -socle then the same would be true of the truncation, since no composition factors are killed under truncation Claim (c) for the remaining cases is trivial 4 Results for p = 3 In characteristic 3 several of the Weyl modules one must consider are non-generic due to the proximity of their highest weight to the upper wall of the lowest p 2 -alcove This leads ultimately to examples of non-rigid tilting modules Another complication is that the G 1 T -socles of two direct summands of the tensor square of the Steinberg module fail to be simple 41 Structure of certain Weyl modules for p = 3 We record the structure of certain Weyl modules needed later The uniserial Weyl modules that turn up in our tensor product decompositions have structure given by (0, 0) = L(0, 0), (1, 0) = L(1, 0), (2, 0) = L(2, 0), (2, 1) = L(2, 1), (2, 2) = L(2, 2), (5, 2) = L(5, 2),

10 C BOWMAN, SR DOTY, AND S MARTIN (1, 1) = [(1, 1), (0, 0)], (3, 0) = [(3, 0), (1, 1)], (4, 0) = [(4, 0), (0, 2)], (3, 1) = [(3, 1), (1, 2)], (5, 0) = [(5, 0), (0, 1)], (5, 1) = [(5, 1), (1, 0)], (3, 2) = [(3, 2), (1, 3), (2, 1)], (6, 0) = [(6, 0), (4, 1), (0, 0)] (4, 2) = [(4, 2), (0, 4), (2, 0)], (6, 2) = [(6, 2), (4, 3), (1, 0), (5, 1)] We note that the structure of (6, 2) is needed only in the Appendix The non-uniserial cases we need have structure (4, 1) = (4,1) (0,3) (0,0) (3,0) (1,1), (4, 3) = (4,3) (1,0) (0,5) (5,1) (1,0), (3, 3) = (3,3) (0,0) (1,4) (4,1) (0,3) (0,0) (3,0) (1,1), (4, 4) = (4,4) (3,3) (1,1) (0,6) (0,3) (0,0) (3,0) (6,0) (1,4) (4,1) (0,0) As for the case p = 2, these structures were obtained by explicit calculations in the hyperalgebra, using GAP to do the calculations 42 Restricted tensor product decompositions for p = 3 The indecomposable decompositions of restricted tensor products for p = 3 is given below We omit any decomposition of the form L(λ) L(µ) where one of λ, µ is zero, and we omit all cases that follow by applying symmetry to a case already listed Proposition Let p = 3 (a) The indecomposable direct summands of tensor products of non-trivial restricted simple SL 3 -modules are as follows: (1) L(1, 0) L(1, 0) T(2, 0) T(0, 1); (2) L(1, 0) L(0, 1) T(1, 1); (3) L(1, 0) L(2, 0) T(3, 0); (4) L(1, 0) L(1, 1) T(2, 1) T(0, 2); (5) L(1, 0) L(0, 2) T(1, 2) T(0, 1); (6) L(1, 0) L(2, 1) T(3, 1) T(2, 0); (7) L(1, 0) L(1, 2) T(2, 2) T(0, 3); (8) L(1, 0) L(2, 2) T(3, 2); (9) L(2, 0) L(2, 0) T(4, 0) T(2, 1); () L(2, 0) L(1, 1) T(3, 1) T(0, 1); (11) L(2, 0) L(0, 2) T(2, 2) T(1, 1); (12) L(2, 0) L(2, 1) T(4, 1) T(2, 2);

11 DECOMPOSITION OF TENSOR PRODUCTS 11 (13) L(2, 0) L(1, 2) T(3, 2) T(0, 2) T(1, 0); (14) L(2, 0) L(2, 2) T(4, 2) T(2, 3); (15) L(1, 1) L(1, 1) T(2, 2) T(0, 0) M; (16) L(1, 1) L(2, 1) T(3, 2) T(4, 0) T(1, 0); (17) L(1, 1) L(2, 2) T(3, 3) T(2, 2); (18) L(2, 1) L(2, 1) T(4, 2) T(5, 0) T(2, 3) T(3, 1); (19) L(2, 1) L(1, 2) T(3, 3) 2T(2, 2) T(1, 1); (20) L(2, 1) L(2, 2) T(4, 3) 2T(3, 2) T(2, 4); (21) L(2, 2) L(2, 2) T(4, 4) T(3, 3) T(5, 2) T(2, 5) 3T(2, 2) Thus the family F is in this case given by the twenty-five tilting modules {T(a, b) : 0 a, b 4} along with the six exceptional modules {T(5, 0), T(0, 5), T(5, 2), T(2, 5), L(1, 1), M} All members of F except L(1, 1) and M are tilting modules (b) The uniserial members of F have the following structure: T(0, 0) = [(0, 0)]; T(1, 0) = [(1, 0)]; T(2, 0) = [(2, 0)]; T(1, 1) = [(0, 0), (1, 1), (0, 0)]; T(2, 1) = [(2, 1)]; T(4, 0) = [(0, 2), (4, 0), (0, 2)]; T(3, 1) = [(1, 2), (3, 1), (1, 2)]; T(2, 2) = [(2, 2)]; T(5, 0) = [(0, 1), (5, 0), (0, 1)]; T(5, 2) = [(5, 2)] The structure of the non-uniserial rigid members of F is given below (symmetric cases omitted): (1,1) (3,0) (0,0) (1,1) (1,1) (3,0) (0,0) (0,3) (1,1) (2,1) (1,3) (2,1) (3,2) (1,3) (2,1) (1,1) (0,3) (0,0) (3,0) (1,1) (4,1) (1,1) (3,0) (0,0) (0,3) (1,1) (2,0) (0,4) (2,0) (4,2) (0,4) all of which are tilting modules excepting the module M (which does not have a highest weight) pictured at the upper right Finally, there are four members of F, namely T(3, 3), T(4, 3), T(3, 4), and T(4, 4), whose structure is not rigid, which are not pictured Analysis of their structure requires other methods (see the Appendix) (c) Each member of F except T(5, 2) = L(5, 2), T(2, 5) = L(2, 5) has simple G 1 T -socle (and head) of highest weight belonging to X 1 Remark The Alperin diagram for T(4, 1) given above is one of several possibilities When a module has a direct sum of two or more copies of the same simple on a given socle layer, there may be more than one diagram The proof of the proposition will be given in First we consider some consequences and look at a few examples (2,0)

12 12 C BOWMAN, SR DOTY, AND S MARTIN Corollary Let p = 3 Given arbitrary dominant weights λ, µ write λ = λ j p j, µ = µ j p j with each λ j, µ j X 1 (a) In the decomposition (113), every direct summand not involving a tensor factor of the form T(5, 2), T(2, 5) is indecomposable (b) L(λ) L(µ) is indecomposable if and only if for each j 0 the unordered pair {λ j, µ j } is one of the cases {(1, 0), (0, 1)], {(1, 0), (2, 0)], {(1, 0), (2, 2)], {(0, 1), (0, 2)], {(0, 1), (2, 2)] or one of λ j, µ j is the zero weight (0, 0) (c) Let m be the maximum j such that at least one of λ j, µ j is non-zero Then L(λ) L(µ) is indecomposable tilting, isomorphic to T(λ + µ), if and only if: (i) for each 0 j m 1, one of λ j, µ j is minuscule and the other is the Steinberg weight (2, 2), and (ii) {λ m, µ m } is one of the cases listed in part (b) Proof The proof is entirely similar to the proof of the corresponding result in the p = 2 case We leave the details to the reader Examples (i) We work out the indecomposable direct summands of L(5, 4) L(4, 5), using information from part (a) of the proposition and following the procedure of Section 11: L(5, 4) L(4, 5) ( L(2, 1) L(1, 1) [1]) ( L(1, 2) L(1, 1) [1]) ( L(2, 1) L(1, 2) ) ( L(1, 1) L(1, 1) ) [1] ( T(3, 3) 2T(2, 2) T(1, 1) ) ( T(2, 2) T(0, 0) M ) [1] ( T(3, 3) T(2, 2) [1]) ( T(3, 3) T(0, 0) [1]) ( T(3, 3) M [1]) 2 ( T(2, 2) T(2, 2) [1]) 2 ( T(2, 2) T(0, 0) [1]) 2 ( T(2, 2) M [1]) ( T(1, 1) T(2, 2) [1]) ( T(1, 1) T(0, 0) [1]) ( T(1, 1) M [1]) T(9, 9) T(3, 3) ( T(3, 3) M [1]) 2T(8, 8) 2T(2, 2) 2 ( T(2, 2) M [1]) T(7, 7) T(1, 1) ( T(1, 1) M [1]) We applied (244) to get the last line of the calculation (ii) Illustrating part (b) of the corollary we have L(3, 1) L(1, 3) L(0, 1) L(1, 0) [1] L(1, 0) L(0, 1) [1] T(1, 1) T(1, 1) [1], which is indecomposable but not tilting (iii) To illustrate part (c) of the corollary we have for instance L(4, 0) L(8, 8) T(12, 8) or L(5, 2) L(5, 4) T(, 6) 43 We now discuss the problem of computing the indecomposable direct summands (and their multiplicities) of L(λ) L(µ) for arbitrary λ, µ X +, in the more difficult case where a direct summand on the right hand side of (113) is not necessarily indecomposable It will be convenient to introduce the notation F 0 for the set F {T(5, 2), T(2, 5)} The Corollary 42(a) says that a direct summand S = j 0 I[j] j in (113) is indecomposable whenever all its tensor factors I j belong to F 0 Consider a summand S = j 0 I[j] j in (113) which is possibly not indecomposable By Corollary 42(a), such a summand must have one or more tensor multiplicands of the form T(5, 2) or T(2, 5) Suppose that in the summand in question I k is T(5, 2) or T(2, 5) We

13 DECOMPOSITION OF TENSOR PRODUCTS 13 use the fact that T(5, 2) = L(5, 2) L(2, 2) L(1, 0) [1], and similarly T(2, 5) = L(2, 5) L(2, 2) L(0, 1) [1] Thus we are forced to consider the possible splitting of L(1, 0) I k+1 or L(0, 1) I k+1 in degree k + 1 There are two cases We consider first the case where I k+1 is not tilting, ie, I k+1 is either L(1, 1) or M So we need to split L(1, 0) L(1, 1), L(0, 1) L(1, 1), L(1, 0) M or L(0, 1) M The first two cases are already covered by Corollary 42(a), so we just need to consider the last two But a simple calculation with characters and consideration of linkage classes shows that (431) L(1, 0) M T(3, 1) T(1, 0) T(4, 0); L(0, 1) M T(1, 3) T(0, 1) T(0, 4) and the summands are once again members of F with restricted socles, so these cases present no problem We are left with the case where I k+1 is tilting then this splitting can be computed since L(1, 0) = T(1, 0) and L(0, 1) = T(0, 1) are tilting, so we are just splitting a tensor product of two tilting modules into a direct sum of indecomposable tilting modules, which can always be done This new decomposition produces only tilting modules in the family F except when I k+1 is one of the following cases: T(5, 0), T(4, 1), T(4, 2), T(5, 2), T(4, 3), and T(4, 4) or one of their symmetric cousins Up to lower order terms which again belong to F, these possibilities, when tensored by L(1, 0) or L(0, 1), produce the new tilting modules (432) T(6, 0), T(5, 1), T(6, 2), T(5, 3), and T(5, 4) and of course their symmetric versions Now by Donkin s tensor product theorem we have a twisted tensor product decomposition for the last three of these, in terms of members of F: (433) T(6, 2) T(3, 2) T(1, 0) [1], T(5, 3) T(2, 3) T(1, 0) [1], T(5, 4) T(2, 4) T(1, 0) [1] Hence, those summands and their symmetric versions present no problem Finally, if T(5, 0) or T(4, 1) is tensored by L(1, 0) then, modulo lower order terms which belong to F, we obtain the new summands T(6, 0) and T(5, 1) which are not members of F and do not admit a twisted tensor product decomposition However, these summands must have simple restricted G 1 T -socles, since they are embedded in T(4, 4) and T(4, 3), respectively This is shown by translation arguments, similar to those in 412 ahead Thus we have proved the following result Proposition Let p = 3 and G = SL 3 (a) Any tensor product of the form L(1, 0) I or L(0, 1) I, where I is an indecomposable tilting module in F, is expressible as a twisted tensor product of modules which are either tilting modules in F 0 or are one of the extra modules T(6, 0), T(5, 1), T(0, 6) or T(1, 5) (b) The extra modules have simple restricted G 1 T -socles

14 14 C BOWMAN, SR DOTY, AND S MARTIN (c) For general λ, µ X +, the indecomposable direct summands of L(λ) L(µ) are all expressible as twisted tensor products of modules from the family F = F 0 {T(6, 0), T(5, 1), T(0, 6), T(1, 5)} = {T(a, b): 0 a, b 4} {T(6, 0), T(5, 1), T(5, 0), T(0, 6), T(1, 5), T(0, 5), L(1, 1), M} Note that all members of F have simple G 1 T -socle of restricted highest weight Example We consider an example where the direct summands on the right hand side of (113) are not all indecomposable: L(2, 2) L(5, 2) L(2, 2) L(2, 2) L(1, 0) [1] ( T(4, 4) T(3, 3) T(5, 2) T(2, 5) 3T(2, 2) ) L(1, 0) [1] T(7, 4) T(6, 3) T(8, 2) T(2, 5) T(5, 5) 3T(5, 2) The second line comes from equation (21) in Proposition 42(a), and to get the last line one applies (244) repeatedly, using Proposition 42(a) again as needed For instance, using equation (1) from Proposition 42(a) we have T(5, 2) L(1, 0) [1] L(2, 2) ( L(1, 0) L(1, 0) ) [1] L(2, 2) ( T(2, 0) T(0, 1) ) [1] T(8, 2) T(2, 5) and using equation (2) from Proposition 42(a) we have T(2, 5) L(1, 0) [1] L(2, 2) ( L(0, 1) L(1, 0) ) [1] L(2, 2) T(1, 1) [1] T(5, 5) 44 We now embark upon the proof of Proposition 42 First we compute the composition factor multiplicities of the restricted tensor products (Recall that χ p (λ) = L(λ) is the formal character of L(λ)) (1) χ p (1, 0) χ p (1, 0) = χ p (2, 0) + χ p (0, 1); (2) χ p (1, 0) χ p (0, 1) = χ p (1, 1) + 2χ p (0, 0); (3) χ p (1, 0) χ p (2, 0) = χ p (3, 0) + 2χ p (1, 1) + χ p (0, 0); (4) χ p (1, 0) χ p (1, 1) = χ p (2, 1) + χ p (0, 2); (5) χ p (1, 0) χ p (0, 2) = χ p (1, 2) + χ p (0, 1); (6) χ p (1, 0) χ p (2, 1) = χ p (3, 1) + 2χ p (1, 2) + χ p (2, 0); (7) χ p (1, 0) χ p (1, 2) = χ p (2, 2) + χ p (0, 3) + 2χ p (1, 1) + χ p (0, 0); (8) χ p (1, 0) χ p (2, 2) = χ p (3, 2) + 2χ p (1, 3) + 3χ p (2, 1); (9) χ p (2, 0) χ p (2, 0) = χ p (4, 0) + χ p (2, 1) + 2χ p (0, 2); () χ p (2, 0) χ p (1, 1) = χ p (3, 1) + 2χ p (1, 2) + χ p (0, 1); (11) χ p (2, 0) χ p (0, 2) = χ p (2, 2) + χ p (1, 1) + 2χ p (0, 0); (12) χ p (2, 0) χ p (2, 1) = χ p (4, 1) + χ p (2, 2) + 2χ p (0, 3) + 2χ p (3, 0) + 4χ p (1, 1) + 2χ p (0, 0); (13) χ p (2, 0) χ p (1, 2) = χ p (3, 2) + 2χ p (1, 3) + 3χ p (2, 1) + χ p (0, 2) + χ p (1, 0); (14) χ p (2, 0) χ p (2, 2) = χ p (4, 2) + χ p (2, 3) + 2χ p (0, 4) + 2χ p (3, 1) + 3χ p (1, 2) + 3χ p (2, 0); (15) χ p (1, 1) χ p (1, 1) = χ p (2, 2) + χ p (0, 3) + χ p (3, 0) + 2χ p (1, 1) + 2χ p (0, 0); (16) χ p (1, 1) χ p (2, 1) = χ p (3, 2) + 2χ p (1, 3) + χ p (4, 0) + 3χ p (2, 1) + 2χ p (0, 2) + χ p (1, 0);

15 DECOMPOSITION OF TENSOR PRODUCTS 15 (17) χ p (1, 1) χ p (2, 2) = χ p (3, 3)+2χ p (1, 4)+2χ p (4, 1)+χ p (2, 2)+4χ p (0, 3)+4χ p (3, 0)+ 6χ p (1, 1) + 5χ p (0, 0); (18) χ p (2, 1) χ p (2, 1) = χ p (4, 2) + χ p (2, 3) + 2χ p (0, 4) + χ p (5, 0) + 3χ p (3, 1) + 5χ p (1, 2) + 3χ p (2, 0) + 2χ p (0, 1); (19) χ p (2, 1) χ p (1, 2) = χ p (3, 3)+2χ p (1, 4)+2χ p (4, 1)+2χ p (2, 2)+4χ p (0, 3)+4χ p (3, 0)+ 7χ p (1, 1) + 7χ p (0, 0); (20) χ p (2, 1) χ p (2, 2) = χ p (4, 3)+χ p (2, 4)+2χ p (0, 5)+2χ p (5, 1)+2χ p (3, 2)+4χ p (1, 3)+ 2χ p (4, 0) + 6χ p (2, 1) + 3χ p (0, 2) + 5χ p (1, 0); (21) χ p (2, 2) χ p (2, 2) = χ p (4, 4) + χ p (2, 5) + 2χ p (0, 6) + χ p (5, 2) + 3χ p (3, 3) + 6χ p (1, 4) + 2χ p (6, 0) + 6χ p (4, 1) + 3χ p (2, 2) + 8χ p (0, 3) + 8χ p (3, 0) + 11χ p (1, 1) + 15χ p (0, 0) It is important to proceed inductively through the cases, so that the structure of smaller tilting modules is available by the time the argument reaches the higher, more complicated, cases We order the cases as listed in part (a) of Proposition 42 In each case, one starts by partitioning the composition factors into blocks This amounts to looking at linkage classes of the highest weights of those composition factors In cases (1) (9), (11) (14) the argument is entirely similar to the arguments already used in characteristic 2, in the proof of Proposition 32 In these cases we know that the tensor product in question is tilting, and it turns out that each linkage class determines an indecomposable direct summand This uses the contravariant self-duality of the tilting modules and the structural information in 41 for the Weyl modules, which forces a lower bound on the composition length of the tilting module in question, and it turns out that this lower bound agrees with the upper bound provided by the linkage class As an example, let us examine the argument in the case (12), for the tensor product L(2, 0) L(2, 1) The linkage classes are {(2, 2)} {(4, 1), (3, 0), (0, 3), (1, 1), (0, 0)} By highest weight considerations, we must have a single copy of T(4, 1) in L(2, 0) L(2, 1) Linkage forces a copy of L(2, 2) = T(2, 2) to split off as well Now T(4, 1) has a submodule isomorphic to (4, 1), so L(1, 1) is contained in its socle By self-duality of T(4, 1), we must have another copy of L(1, 1) in the top of T(4, 1), so we are forced to put a copy of (1, 1) at the top of T(4, 1) Looking at the structure of (4, 1) in 516, we see that T(4, 1) must also have at least one copy of (3, 0) and (0, 3) in its -filtration At this point we are finished, since this accounts for all available composition factors (with their multiplicities) from the linkage class, so we conclude that L(2, 0) L(2, 1) T(4, 1) T(2, 2) The structure of T(4, 1) is nearly forced, because of its self-duality, the fact that all the Ext groups between simple factors is known, and the fact that T(4, 1) must have both and -filtrations In 412 we will see that T(0, 3) is isomorphic to a submodule of T(4, 1), which finishes the determination of the structure of T(4, 1) 45 In case (), one cannot immediately conclude that L(2, 0) L(1, 1) is tilting since L(1, 1) is not tilting, so we must proceed differently However, we observe the following, which immediately implies that in fact our tensor product is tilting Lemma Let V be a simple Weyl module and let (λ) be a Weyl module of highest weight λ If the composition factors of V rad (λ) and V L(λ) lie in disjoint blocks, then V L(λ) is tilting

16 16 C BOWMAN, SR DOTY, AND S MARTIN Proof V (λ) has a -filtration, by the Wang Donkin Mathieu result (see [14, II421]) Now as V rad (λ) and V L(λ) have no common linkage classes there can be no non-trivial ( extensions ) ( between ) these modules, by the linkage principle Thus V (λ) = V rad (λ) V L(λ) As V (λ) has a -filtration this implies V L(λ) does also As it is the tensor product of two simple (therefore contravariantly self dual) modules it is itself contravariantly self dual and so has a -filtration Therefore it is tilting Now we may proceed as usual Looking at the character of L(2, 0) L(1, 1) we find that there are two linkage classes for the highest weights of the composition factors, namely {(0, 1)} and {(3, 1), (1, 2)} Since the multiplicity of L(0, 1) is 1, it must give a simple tilting summand T(0, 1) Now T(3, 1) must be a summand by highest weight consideration, and the usual argument forces it to have at least composition length three, which forces equality of the upper and lower bounds, so the structure is T(3, 1) = [(1, 2), (3, 1), (1, 2)] and we have L(2, 0) L(1, 1) T(3, 1) T(0, 1) This takes care of case () in our list Case (16) follows similarly, making use again of the above lemma to conclude that L(1, 1) L(2, 1) is tilting We note that at this stage we may assume that the structure of T(3, 2) and T(4, 0) are already known, since they come up in the earlier cases (8), (9) So one easily concludes from this and the linkage classes that L(1, 1) L(2, 1) T(3, 2) T(4, 0) T(1, 0) 46 We now consider case (15) Since L(1, 1) is not tilting, it is unclear whether or not L(1, 1) L(1, 1) is tilting In fact it is not, and analysis of this case is more difficult First, looking at the character and the linkage classes (there are two) we observe that a copy of the Steinberg module T(2, 2) = L(2, 2) splits off as a direct summand The remaining composition factors of the tensor product all lie in the same linkage class, but it turns out that a copy of the trivial module splits off, as we show below From properties of duals and previous calculations it follows that (1) dim K Hom G (L(0, 0), L(1, 1) L(1, 1)) = dim K Hom G (L(0, 0) L(1, 1), L(1, 1)) = dim K Hom G (L(1, 1), L(1, 1)) = 1; (2) dim K Hom G (T(1, 1), L(1, 1) L(1, 1)) = dim K Hom G (L(1, 0) L(0, 1), L(1, 1) L(1, 1)) = dim K Hom G (L(1, 1) L(0, 1), L(1, 1) L(0, 1)) = dim K Hom G (L(1, 2) L(2, 0), L(1, 2) L(2, 0)) = 2; (3) dim K Hom G (L(3, 0), L(1, 1) L(1, 1)) = dim K Hom G (L(1, 1) L(3, 0), L(1, 1)) = dim K Hom G (L(4, 1), L(1, 1)) = 0 and by symmetry, an equality similar to (3) holds, in which (3, 0) is replaced by (0, 3) By (1) and (2) and considerations of the three homomorphic images of the module T(1, 1) we see that L(1, 1) L(1, 1) has either: (a) a copy of L(0, 0) in the first socle layer, a copy of L(1, 1) in the second socle layer, and a copy of L(0, 0) in the third socle layer, or

17 DECOMPOSITION OF TENSOR PRODUCTS 17 (b) a copy of L(0, 0) and L(1, 1) in the first socle layer and a copy of L(0, 0) in the second socle layer In fact, by known Ext 1 information and self-duality possibility (a) is ruled out, and one sees that L(1, 1) L(1, 1) has two simple direct summands, namely L(2, 2) and L(0, 0), and the remaining indecomposable summand must have the diagram (1,1) (3,0) (0,0) (0,3) (1,1) which proves that L(1, 1) L(1, 1) L(2, 2) L(0, 0) M, as claimed 47 There are just five cases remaining in the proof of Proposition 42, namely cases (17) (21) We now consider case (17) The module L(1, 1) L(2, 2) is tilting by Lemma 45, so by highest weight considerations T(3, 3) is a direct summand This is also justified by Pillen s Theorem (see 21) The character of T(3, 3) may be computed by (222), which shows that it has a -filtration with -factors isomorphic to (3, 3), (4, 1), (1, 4), (3, 0), (0, 3), (1, 1) each occurring with multiplicity one This accounts for all the composition factors appearing in the character of L(1, 1) L(2, 2), except for one copy of the Steinberg module T(2, 2) = L(2, 2) Hence we conclude that L(1, 1) L(2, 2) T(3, 3) T(2, 2) 48 L(2, 1) L(2, 1) is tilting since L(2, 1) is, so by highest weight considerations a copy of T(4, 2) splits off as a direct summand The structure of T(4, 2) was determined in a previous case of the proof Subtracting its character from the character of L(2, 1) L(2, 1), we see that the highest weight of what remains is (5, 0), so a copy of T(5, 0) must split off as well The linkage class of (5, 0) contains only two weights {(5, 0), (0, 1)} and from this and the known structure of the Weyl modules it follows easily that T(5, 0) is uniserial with structure T(5, 0) = [(0, 1), (5, 0), (0, 1)] Now highest weight and character considerations force the remaining summands to be one copy of T(2, 3) and one copy of T(3, 1) Hence L(2, 1) L(2, 1) T(4, 2) T(5, 0) T(2, 3) T(3, 1) We note we can assume that T(3, 1) and T(2, 3) are known at this point, since they arise in earlier cases of the proof (Actually, to be precise T(2, 3) doesn t arise in any earlier case, but its symmetric cousin T(3, 2) does) 49 L(2, 1) L(1, 2) is tilting since both L(2, 1) and L(1, 2) are, so by highest weight considerations a copy of T(3, 3) splits off as a direct summand The character of T(3, 3) was computed already in 47, so by character considerations one easily deduces that L(2, 1) L(1, 2) T(3, 3) 2T(2, 2) T(1, 1) Of course, the character of T(1, 1) is already known by an earlier case of the proof

18 18 C BOWMAN, SR DOTY, AND S MARTIN 4 L(2, 1) L(2, 2) is tilting since both L(2, 1) and L(2, 2) are, so by highest weight considerations a copy of T(4, 3) splits off as a direct summand From (222) we compute its -factors to be (4, 3), (5, 1), (0, 5), (1, 0) This also follows from the translation principle applied to T(3, 3) One sees also that T(4, 3) has simple socle of highest weight (1, 0) by arguments similar to those in 47 From character computations one now shows that L(2, 1) L(2, 2) T(4, 3) 2T(3, 2) T(2, 4) The structure of T(3, 2) is available by a previous case of the proof, and the structure of T(2, 4) follows by symmetry from that of T(4, 2), again a previous case 411 L(2, 2) L(2, 2) is tilting since L(2, 2) is, so by highest weight considerations a copy of T(4, 4) must split off as a direct summand The -factor multiplicities of T(4, 4) are computed by (222) to be (4, 4), (6, 0), (0, 6), (3, 3), (4, 1), (1, 4), (1, 1), (0, 0) each of multiplicity one From this, using the character of L(2, 2) L(2, 2) it follows by highest weight considerations, after subtracting the character of T(4, 4), that a copy of T(3, 3) must also split off as a direct summand Then it easily follows that L(2, 2) L(2, 2) T(4, 4) T(3, 3) T(5, 2) T(2, 5) 3T(2, 2) where T(5, 2) = L(5, 2), T(2, 5) = L(2, 5), and T(2, 2) = L(2, 2) At this point the proof of Proposition 42(a), (b) is complete 412 It remains to prove the claim in part (c) of Proposition 42 It is known that Donkin s conjecture holds for SL 3, so T((p 1)ρ + λ) is as a G 1 T -module isomorphic to Q 1 ((p 1)ρ + w 0 λ) for any λ X 1 Thus T(a, b) has simple G 1 T -socle of restricted highest weight, for any 2 a, b 4 Moreover, the claim is true of T(0, 0), T(1, 0), T(2, 0), T(2, 1), L(1, 1) and their symmetric counterparts, since these are all simple G-modules of restricted highest weight For λ = (1, 1) and (5, 0) one easily checks by direct computation that (λ), which is a non-split extension between two simple G-modules, remains non-split upon restriction to G 1 T It then follows that T(λ) has simple G 1 T -socle of restricted highest weight in each case For λ = (4, 0) and (3, 1) one could argue as in the preceding paragraph, or restrict to an appropriate Levi subgroup, as in the last paragraph of 33 The remaining cases, up to symmetry, are T(3, 0), T(4, 1), and M We apply the translation principle [14, IIE11] Observe (from their structure) that T(0, 2) embeds in T(4, 0), which in turn embeds in T(2, 4) Picking λ = (0, 0) and µ = ( 1, 1) in the closure of the bottom alcove, observe that applying the (exact) functor T λ µ to these embeddings, we obtain embeddings of T(0, 3) in T(4, 1), and T(4, 1) in T(3, 3) Since T(3, 3) has simple G 1 T -socle of restricted highest weight, it follows that the same holds for T(0, 3) and T(4, 1) The cases T(3, 0) and T(1, 4) are treated by the symmetric argument Finally, we observe that dim K Hom G1 T (L(0, 0), L(1, 1) L(1, 1)) = 1, by a calculation similar to 46(1) This, along with 46, shows that M remains indecomposable on restriction to G 1 T, with socle and head isomorphic to L(1, 1), and with 7 copies of L(0, 0) in the middle Loewy layer The proof of Proposition 42 is complete

19 DECOMPOSITION OF TENSOR PRODUCTS Discussion We now discuss the remaining issue in characteristic 3: the structure of the tilting modules T(λ) for λ = (3, 3), (4, 3), (34), and (4, 4) These tilting modules are in fact S-modules for the Schur algebra S = S K (3, r) in degree r = 9,, 11, 12, respectively (See [, 17] for background on Schur algebras) Thus, in order to study the structure of T(λ) one may employ techniques from the theory of finite dimensional quasi-hereditary algebras Now the simplest cases (in terms of number of composition factors) are T(4, 3) for S(3, ) and T(3, 4) for S(3, 11) As these modules are symmetric, it makes sense to focus on the smaller Schur algebra S(3, ) and thus T(4, 3) In fact, it is enough to understand the block A of S(3, ) consisting of the six weights (, 0), (6, 2), (4, 3), (5, 1), (0, 5), and (1, 0) (It is easily seen that this is a complete linkage class of dominant weights in S(3, ), for instance by drawing the alcove diagrams) To construct T(4, 3) we must glue together the -factors in a way that results in a contravariantly self-dual module Looking at the diagrams in Figure 1 below picturing (1,0) (0,5) (5,1) (1,0) (1,0) (4,3) (1,0) (0,5) (5,1) (1,0) Figure 1 Weyl filtration factors of T(4, 3) the various Weyl modules in the filtration, we see that it is impossible to do this in a rigid way There are three copies of L(1, 0) above the middle factor L(4, 3) and only two below Thus, there must be two copies of L(1, 0) lying immediately above L(4, 3) when viewing the radical series, and two copies lying immediately below L(4, 3) when viewing the socle series This implies that T(4, 3) is not rigid To understand the structure of T(4, 3) one may apply Gabriel s theorem to find a quiver and relations presentation for the basic algebra of the block A, or an appropriate quasi-hereditary quotient thereof This is carried out in the Appendix The other cases could be treated similarly Note that none of T(4, 3), T(3, 4), T(3, 3), or T(4, 4) is projective as an S-module, because if so, the reciprocity law (P (λ): (µ)) = [ (µ): L(λ)] (see eg [4, Prop A22]) would be violated References [1] JL Alperin, Diagrams for modules, J Pure Appl Algebra 16 (1980), [2] HH Andersen and M Kaneda, Rigidity of tilting modules, preprint [3] S Donkin, On tilting modules for algebraic groups, Math Z 212 (1993), [4] S Donkin, The q-schur algebra, London Math Soc Lecture Note Ser, 253, Cambridge Univ Press 1998